2/6/17

Today I finished the first unit of my course on deep learning. The last phase of the first project involved using a process called parameter tuning to improve the accuracy of the neural network. Parameter tuning works like it sounds; you change the parameters of the neural network (like batch size, number of training epochs, etc.) to optimize its performance. There are tools that you can use in Python to find the most effective combination of parameters for a neural network, but they take a loooong time to train so I couldn't actually train the network in the span of this class period. The lecturer said that it took several hours for his model to train, so its the kind of thing you leave on overnight. After the lecture concluded I experimented with different combinations of parameters to see how the different combinations affected the network's accuracy. The "homework" assignment is to get above 86% accuracy through experimentation (you get a "gold medal" if you do so). Oops forgot to submit this in during class.

Comments

Popular Posts